dear editor and anonymous referee #2 · 1km buffer zone established for each sample in this study,...
TRANSCRIPT
Replies to the comments and suggestions Dear editor and Anonymous Referee #2:
Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled” Recognition
of spatial framework for water quality and its relation with land use/cover types from a new perspective: A case
study of Jinghe Oasis in Xinjiang, China”. Those comments are all valuable and very helpful for revising and
improving our paper, as well as important guiding sense to our researches. The authors have studied comments
carefully and have made correction which we hope meet with approval. Revised portions are marked in red in the
paper and authors have tried our best to revise the manuscript based on their suggestions. The point-by-point
responses to each of the comments are presented as follows.
Anonymous Referee #2
This manuscript conducted a very detailed work on investigating the water quality and its relation with land
use/cover types. While the data and samples have been processed carefully and rigorously, the results look
reasonable and trustful. However, I have several major concerns regarding to this work.
Reply: Thank you for your positive evaluation and encouragement. We have carefully revised the paper according
to your comments and suggestions. The responses to the specific comments are presented below
Q: There are two key words in the title that I cannot find strong evidence from the context to support, the “spatial”
and “a new perspective”. I cannot see how the authors make their findings ‘spatial’, since there is nothing that has
been shown as a map, except for the study area (Figure 1) and land cover (Figure 6). From the Introduction, I also
cannot see why the authors labeled their work as “a new perspective”? It is new because of what? This should be
made clear in the Introduction.
Reply: Thanks to Reviewers for their suggestions in improving the manuscript, based on reviewer 1, the title is
revised; ‘spatial’ is changed “spatial pattern”. “a new perspective” is deleted by authors.
The title has changed into “Recognizing spatial patterns in water quality and their relation with land use/cover
types: A case study of the Jinghe Oasis in Xinjiang, China”
Q: I am also unclear about the role that remote sensing data played in this study. What was it for, only for
revealing the land use/cover types of each sample? How was it used? It was said there was a 1km buffer zone
established for each sample. Then how? And why 1km? Besides, in line 246, there is the accuracy of land
use/cover classification. I am wondering how these accuracy numbers were derived? Using what as reference
(ground truth)?
Reply: In order to extract the area of land use/cover types, therefore, remote sensing data is selected in this study.
1km buffer zone established for each sample in this study, based on most research (Wang et al., 2016; wang
et al., 2017) and regional natural situation, Wang et al., (2016) reports a correlation between croplands areas and
water quality in rivers, but this correlation isn’t significant in 500 m. Wang et al., (2017) reports a significant
correlation between land use and cover areas and water quality in kilometer scale in rivers. Because the study area
is located in the arid areas of Xinjiang, desert dust and salt dust are major environmental hazards in this study area.
Desert dust and salt dust seriously affect the atmosphere and water quality and accelerate the degradation of
vegetation and threaten ecological security in the oasis. However, with economic development of and population
growth, human improvement of desert land and saline land increases the area of croplands (Yu et al., 2017), and
small changed in landscape. Therefore, 1km buffer zone established for each sample in this study.
The accuracy of land use/cover classification is required by confusion matrix, historical data, high-resolution
Google Earth images, and field survey data, selected to verify that more than 100 pixels of each land cover type
were used for the training data, and the confusion matrix to verify the classification results. From the classification
results obtained in May and October 2015 (Figure 6), precision has increased to 89.9750% and 86.2848%, and the
kappa coefficients are 0.8681and 0.8184, respectively by confusion matrix (Table 2). Please see the article on page
6-7, lines to246-250 in revised manuscript with obviously marked as follows:
In this study, historical data, high-resolution Google Earth images, and field survey data were selected to
verify that more than 100 pixels of each land cover types were used for the training data, and the confusion matrix
was used to verify the classification results. For the classification results obtained in May and October 2015
(Figure 6), precision increased to 89.9750% and 86.2848%, respectively, and the kappa coefficients were 0.8681
and 0.8184, respectively, based on the confusion matrix (Table 2),
Table 2 The calculation of a confusion matrix by a maximum likelihood supervised classification
LULC Water body Saline land Farmland Forest
grassland
Other
land Total
User’s
accuracy
(%)
Water body 144 0 0 0 0 144 100
Saline land 0 77 0 0 16 93 82.79
Farmland 0 36 101 0 0 137 73.72
Forest-Grass land 0 36 0 101 0 137 73.72
Other land types 1 0 0 0 87 88 98.96
Total 145 149 101 101 103 Overall=89.9750%
May
Producer’s accuracy (%) 99.31 51.67 100 100 84.46 Kappa=0.8681
Water body 144 0 0 0 0 144 100
Saline land 0 57 0 0 26 83 86.67
Farmland 0 16 101 0 0 117 86.32
Forest-Grass land 4 16 0 101 0 117 86.32
Other land types 0 0 0 0 77 77 100
Total 148 89 101 101 103 Overall=86.2848%
October
Producer’s accuracy (%) 97.29 64 100 100 74.75 Kappa=0.8184
Reference
Wang XP, Zhang F, Li XH, Cao C, Guo M. 2017. Correlation analysis between the spatial characteristics of land
use/cover-landscape pattern and surface-water quality in the Ebinur Lake area. Acta Ecologica sinica, 2017,
37(22):7438~7452
Wang J, Zhang F, Zhang Y, Ren Y, Yu HY, 2016. Correlation between the spatial water quality and land use/cover
in the Ebinur Lake area Acta Ecologica sinica, 2016,36(24):7971~7980.
Yu, H., Zhang, F., Kung, H. T., Johnson, V. C., Bane, C. S., & Wang, J., et al. 2017. Analysis of land cover and
landscape change patterns in ebinur lake wetland national nature reserve, china from 1972 to 2013. Wetlands
Ecology & Management(3), 1-19.
Q: The grammar and usage have pervasive problems. It needs to be comprehensively edited by someone with
strong English language skills before publication. There are some words, such as ‘framework’ in the title, and
‘layers’ in line 19, that are very confused to me. I have no idea what they are referring to.
Reply: Thanks to Reviewers for their suggestions in improving the manuscript, based on reviver 1, the title is
revised, ‘spatial framework’ is changed “”spatial pattern”. But then Professor Hsiang-te Kung (University of
Memphis) has helped us improve the grammar and edit the manuscript carefully. Please see the revised version of
the paper.
Q: Line 132, Davies-Bouldin index just comes out suddenly. What is this? What is the reference for it? What do
you mean by “Through the K-means :”? These content needs to be rewritten.
Reply: The Davies-Bouldin clustering index was used to determine the optimal number of the clusters for a
dataset. The sections of about Davies-Bouldin clustering index is rewritten by authors. Please see the article on
page 4, lines to 129-133 in revised manuscript with obviously marked. As follows:
The Davies-Bouldin clustering index was used to determine the optimal number of clusters for the dataset
(An et al., 2016; Park et al., 2014). The lower the Davies–Bouldin index value is, the better the clusters are
differentiated. The K-means cluster analysis was combined with the Davies–Bouldin index (DBI) to select the
clustering number. (Zhou et al., 2016).
Reference
An, Y., Zou, Z., & Li, R. 2016. Descriptive characteristics of surface water quality in hong kong by a
self-organising map. International Journal of Environmental Research & Public Health, 13(1):115.
Park, Y. S., Kwon, Y. S., Hwang, S. J., & Park, S. 2014. Characterizing effects of landscape and morphometric
factors on water quality of reservoirs using a self-organizing map. Environmental Modelling & Software,
55(5):214-221.
Q: What is the DCA gradient axis in line 163? Why do you say “Therefore, the redundancy analysis (RDA)
method was applied to”. I cannot see connections between the sentences before the word ‘Therefore’ and those
after it.
Reply: The section is revised by authors. Please see the article on page 4, lines to 161-165 in revised manuscript
with obviously marked. As follows:
The results showed that the DCA gradient shaft length was less than 3. Based on the results of Wang et al
(2017), when the DCA gradient shaft length is less than 3, the redundancy analysis (RDA) method can explore the
relationship; therefore, the redundancy analysis (RDA) method was applied to determine the influence trend of
land use/cover changes on the water quality within the Ebinur Lake buffer area.
Reference
Wang XP, Zhang F, Li XH, Cao C, Guo M. 2017. Correlation analysis between the spatial characteristics of land
use/cover-landscape pattern and surface-water quality in the Ebinur Lake area. Acta Ecologica sinica
2017,37(22):7438~7452
Q: Sentence in line184-185 needs to be rewritten.
Reply: Thanks to Reviewers for their suggestions in improving the manuscript, the sentence is rewritten, Please
see the article on page 5, lines to 183-184 in revised manuscript with obviously marked. As follows:
Six clusters were formed according to the DBI, where minimal value is at six clusters as Figure 3a.
In addition, authors have revised the figures and tables as well as words expression in the entire manuscript,
please see the text.
Authors tried our best to improve the manuscript and made some changes in the manuscript. Authors
appreciate for Editors/Reviewers’ warm work earnestly, and hope that the correction will meet with approval.
Once again, thank you very much for your comments and suggestions.
All in all, if you have any questions about our paper, please contact with me as follow address:
E-mail:[email protected]
Thanks very much.
Best wishes and warmly regards for you.
Sincerely yours Fei ZHANG
11th, Jan., 2017
Recognizing spatial patterns in water quality and their relation with 1
land use/cover types: A case study of the Jinghe Oasis in Xinjiang, 2
China 3
Fei ZHANG 1,2,3 Juan WANG 4 Xiaoping WANG 1,2,3 4 5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
1. College of Resources and Environment Science, Xinjiang University, Urumqi, Xinjiang 830046
2. Key Laboratory of Oasis Ecology, Xinjiang University, Urumqi, Xinjiang 830046
3. Key Laboratory of Xinjiang wisdom city and environment modeling, Urumqi, Xinjiang 830046
4. College of Geography and Remote Sensing Science, Beijing Normal University, 100875, Beijing
Abstract: To understand the relation between spatial water quality patterns and changes in land use/cover types in the Jinghe
Oasis, this study divided 47 water sampling sites, which were measured in May and October 2015, into 6 cluster layers using
the self-organizing map (SOM) method based on non-hierarchical k-means classification. Next, it determined the distribution
characteristics of water quality indices for the time sequence. The water quality indices included the chemical oxygen
demand (COD), biological oxygen demand (BOD), suspended solids (SS), total phosphorus (TP), total nitrogen (TN),
ammonia nitrogen (NH3−N), chromaticity (SD), and turbidity (NUT). Based on the results, we collected data on the changes
in the farmland, forest-grass land, water body, salinized land, and other land types during the wet and dry seasons and
combined these data with the classification results of the GF-1 remote sensing satellite data obtained in May and October
2015. We then discussed the influences of land use/cover type on water quality for different layers and seasons. The results
indicate that Clusters 1 to 3 included monitoring samples from the wet season (May 2015), whereas Clusters 4 to 6 included
monitoring samples from the dry season (October 2015). In general, the COD, SS, NUT, TN, and NH3−N values were high
in Clusters 1 and 2. The SD values for Clusters 1, 4, and 6 were high. Moreover, high BOD and TP values were mainly
concentrated in Clusters 4 and 6. In the discussion on the relation between the different water quality parameters and land
use/cover type changes, we determined that farmland, forest-grassland, and salinized land significantly influenced the water
quality parameters in the Jinghe Oasis. In Clusters 1, 2, and 6, the size of the water area, to a certain extent, also influenced
changes in the water quality parameters. In addition, the influences of various land use/cover types on the water quality
parameters in the research zone during the different seasons exhibited the following descending order of magnitude:
farmland → forest-grass land → salinized land → water body → others. Moreover, their influences were lower during the
wet season than the dry season. In conclusion, developing research on the relation between the spatial framework of the
water quality in the Jinghe Oasis and land use/cover type changes is significant for the time sequence distribution of water
quality in arid regions from both theoretical and practical perspectives.
Key words: SOM; Water quality spatial distribution; Land use/cover; Correlation analysis; GIS
0 Introduction
Water quality is of great importance to the study of water resources in arid regions. Accurate
information on the spatial distribution of surface water quality is imperative for assessing environmental
monitoring, land-surface water management and watershed changes (NRC, 2008; Sun et al., 2012). Land
use/cover changes in drainage basins significantly influence the water quality of rivers, lakes, river mouths,
and coastal areas (Huang et al., 2013a; Bu et al., 2014; Hur et al., 2014). Surface water resources, through
runoff or infiltration, will always carry a large amount of pollutants (Swaney et al., 2012). Therefore, the
spatial allocation of land use and land cover changes in drainage basins frequently influences or even
Corresponding author: Zhangfei Tel: +86 13579925126 E-mail address: [email protected]
1
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
endangers water quality through non-point source pollution (Swaney et al., 2012). However, the regional
differences and complexity of land use/cover types result in various relations between land use/cover and
water quality in different regions (Yang et al., 2016). Therefore, it is very important to explore the
relationship between land use/cover types and water quality for the development and management of the
basin. (Uuemaa et al., 2005; Xiao et al., 2007; Wan et al., 2014). At present, numerous scholars have
extensively applied statistical methods to determine the mutual relations between land use/cover changes
and water quality in various research zones (Céréghino et al., 2009; Bierman et al., 2011; Huang et
al.,2013b). These methods include correlation analysis (Lee et al., 2009; Li et al., 2015), multiple
regression (Park et al.,2014), and redundancy analysis (De et al.,2008; Shen et al.,2015).
A self-organizing map (SOM) is a type of artificial neural network algorithm; it is a self-organizing
and self-learning network visual method that can express multi-dimensional spatial data in
low-dimensional points through non-linear mapping (Kohonen., 2001). A SOM is an all-purpose
classification tool that can connect samples with variables (Kohonen., 2013; Zhou et al., 2016). In recent
years, SOMs have become increasingly popular in environmental research because of their capacity to
address non-linear relations. Kalteh (2008) and Céréghino (2009) discussed the application of the SOM
method in environmental science, particularly in water resource classification. Chon (2011) evaluated the
application of SOM technology in the field of ecology. The high-dimensional, non-linear, and uncertain
features of water quality monitoring data result in a certain complexity during the analysis and evaluation
of surface water quality data. Therefore, data mining and the modern mode recognition method have been
introduced to analyse and explain water quality monitoring data, which can, to a certain extent, offset the
deficiency of the traditional method (Li et al.,2013). The Jinghe Oasis, which comprises an oasis and a
desert, is a typical mountainous zone in an arid region and an important part of the northern slope of the
Tianshan Mountains. Under the influences of the drainage basin climate and human activities, the pollution
of the regional ecological environment, which results from agricultural and domestic wastewater sources
around the Jinghe Oasis that are directly discharged or discharged through the river, has become an urgent
problem related to sustainable socioeconomic development in Xinjiang. Therefore, a typical section of the
Jinghe Oasis in the plain area of the arid region was selected as the research object of this study. The SOM
method was applied to recognize the spatial distribution of water quality in the Jinghe Oasis. Based on the
result, this study offers a tentative exploration of the relation between water quality and land use/cover
changes in different clusters and provides new insights on controlling, managing, and protecting the
ecological environment in the Jinghe Oasis.
The main objectives of this study were to (1) analyse the spatial framework of water quality using the
self-organizing map (SOM) method based on non-hierarchical k-means classification; (2) explore the
relationship between the water quality parameters and land use/cover types in different clusters; and (3)
analyse the relationship between water quality parameters and land use/cover types in different stages.
1 Materials and methods
1.1 Study area The Jinghe Oasis is in the centre of Eurasia in the northwest Xinjiang Uygur Autonomous Region at
44°02′~45°10′N and 81°46′~83°51′E. The Jinghe Oasis is composed of wetland and desert oasis
vegetation and wildlife and is a national desert ecological reserve. The study area has a unique wetland
ecological environment, and it has been listed as the Xinjiang Uygur autonomous region "wetland nature
reserve". The region has 385 kinds of desert plants, approximately 64% of the vast amount of desert plants
2
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
in China. The Jinghe Oasis was once fed by 12 river branches belonging to three major river systems, the
Bortala River, Jing River and Kuytun River, which were mainly rivers connected with Ebinur Lake. Due to
natural environmental changes and human activities (i.e., modern oasis agricultural development), many
rivers gradually lost their hydraulic connections with Ebinur Lake, and only Bortala River and Jing River
currently supply water to the lake. The climate in the Jinghe Oasis is a typical continental arid climate,
with an annual average temperature of 7.36 °C, an average precipitation of 100~200 mm, and an average
evaporation of 1500~2000 mm (Zhang et al., 2015). In recent years, under the dual influence of natural and
human factors, the water resources of the Jinghe Oasis have degraded seriously, causing an extreme
decrease in the natural oasis and water area, desertification of the land, salinization of the farmland, serious
grassland degradation, and water quality salinization (Jilil et al.,2002). At the same time, under the effect of
the strong winds in Alashankou, the region has become a main source of dust; this affects the ecological
environment of northern Xinjiang. This study, by using actual data mining, established the research area
shown in Figure 1.
Fig. 1 Location of the study area
1.2 Data acquisition and processing
(1) We applied GF-1 remote sensing images obtained in May and October 2015 as the data sources
(see http://www.cresda.com/CN/). These images were not influenced by clouds, fog, or snow cover, and
their quality was good. We conducted radiation and orthographic corrections for the remote sensing image
data combined with 1:50,000 scale digital elevation model (DEM) data. We established five land use/cover
types by using the Environment for Visualizing Images software (ENVI Version 5.0), namely, farmland,
forest-grassland, water body, salinized land, and others, based on the actual conditions of the research zone.
Finally, we generated a vector data map of the land use/cover types for two stages of the research zones.
(2) The pillar industries in the Jinghe Oasis include salt production and Artemia breeding. No heavy
industry is present; thus, point source pollution from industrial wastewater was not considered in the
research zone. Research samples were collected from agricultural land in Jinghe County and Tuotuo
Village, which surround Ebinur Lake, a national ecological zone in Ebinur Lake Bird Isle, and the Ganjia
Lake Haloxylon natural conservation area. In total, we collected 47 water samples, with 23 collected in
May and 24 collected in October 2015. The monitoring indices used included chemical oxygen demand
(COD), five-day biological oxygen demand (BOD5), suspended solids (SS), total phosphorus (TP), total
nitrogen (TN), ammonia nitrogen (NH3−N), chromaticity (SD), and turbidity (NUT). All polyethylene
bottles were used to store the samples. The bottles were cleaned, dried, and sealed with deionized water
before sampling. The samples were taken to the laboratory for the measurements and analyses after
collection. We applied dichromate titration, dilution and inoculation, gravimetry, ammonium molybdate
spectrophotometry, alkaline potassium persulfate decomposition UV spectrophotometry, and Nessler
reagent spectrophotometry to measure the COD, BOD5, SS, TP, TN, and NH3−N, respectively. The
analyses of all the samples were entrusted to and completed by Urumqi Jincheng Measurement Technology
Co., Ltd.
1.3 Recognition of water quality spatial characteristics based on the SOM method with non-hierarchical k-means classification
At present, classifications based on the SOM neural network are mainly unsupervised and applied to
3
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
fault analyses, text clustering, and water quality evaluations. The method, which does not require a
consistent data distribution, is simple and can address detailed information without the influence of minor
local problems. The results are the distribution features of the input mode and topological structures (Li et
al.,2010). A typical SOM network generally consists of input and output clusters. All the nerve cells in the
input cluster and the weight vectors in the output cluster are connected and classified as typed data using
the SOM via a learning process. Accordingly, the k-means algorithm is applied to keep each cluster
compact and to separate the clusters from each other as much as possible. The Davies-Bouldin clustering
index was used to determine the optimal number of clusters for the dataset (An et al., 2016; Park et al.,
2014). The lower the Davies–Bouldin index value is, the better the clusters are differentiated. The K-means
cluster analysis was combined with the Davies–Bouldin index (DBI) to select the clustering number. (Zhou
et al., 2016).
The SOM method based on non-hierarchical k-means classification was applied to the spatial
framework of the water quality in the research zone by implementing the following steps. (1) We input the
water sample data for clustering from May and October 2015 to the SOM network. We applied the
topological values for calculating the network size to select the quantity of nerve cells and determine the
output results based on the minimum values of the Quantization Error (QE) and Topological Error (TE).
The QE was used to determine the capacity of the established neural network in distinguishing the original
input data, whereas the TE was used to measure the neural network quality, i.e., to evaluate whether the
network was applicable for training (Kohonen, 2001). After determining the network size, we conducted
network training and obtained a set of weight values. (2) The weight value obtained from the SOM
clustering results was considered the initial cluster centre, and the k-means algorithm was initialized to
execute this algorithm and, combined with the DBI index, select the clustering number. This clustering
combination algorithm maintains the self-organizing features of the SOM network, inherits the high
efficiency of the k-means algorithm, and offsets the poor clustering effects that result from the excessive
convergence time of the SOM network and the inappropriate selection of the initial clustering centre for
the k-means algorithm. The SOM requires a SOM toolbox and some basic functions in Matrix Laboratory
(MATLAB) (Zhang,2015). This study used MATLAB 2013a as the calculation platform.
1.4 Spatial analysis of the influences of land use/cover change on water quality
As an artificial system disturbance, land use/cover type is the second major boundary condition that
directly or indirectly influences hydrologic processes and exerts a considerable effect on the drainage water
environment. First, we obtained information on the land use/cover types within the 1 km buffer zone of the
water quality sampling points in the research area using the spatial analysis function of ArcGIS 9.3. Based
on the results, we then discussed and analysed the correlation between water quality and land use/cover
type changes at different levels and periods. For different levels, we established the correlation between
water quality and land use/cover types in each layer and discussed the influence of land use/cover type
changes on water quality. For different periods, analysed the land use/cover type information and eight
types of water quality indices during the dry and wet seasons. The land use/cover type information and
eight water quality indices were imported into Canoco 4.5 (Ter Braak and Smilauer, 2002) to test the DCA
gradient axis. The results showed that the DCA gradient shaft length was less than 3. Based on the results
of Wang et al (2017), when the DCA gradient shaft length is less than 3, the redundancy analysis (RDA)
method can explore the relationship; therefore, the redundancy analysis (RDA) method was applied to
determine the influence trend of land use/cover changes on the water quality within the Ebinur Lake buffer
area. This method indicates the contribution rate of a single land use/cover variable on water quality and
4
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
can directly demonstrate the correlation between land use/cover type and water quality parameters via a 2D
ordination graph. The methodology is explained in the following section, and a conceptual flow chart
describing the methodology is shown in Figure 2.
Fig. 2 Conceptual model of the methodology
2 Results and analysis
2.1 Spatial framework of water quality in the Jinghe Oasis
Regarding the network structure selection, neural networks with a more complicated structure
generally have a better capability to address complicated non-linear problems but require a longer training
time (Kohonen, 2013). Increasing the number of water quality indices can provide more abundant
information; however, the correlations among indices will also increase. The topological values were
selected to determine the grid size in this study, and the k-means clustering method was adopted to obtain
the results. Overall, after the standard processing of the water quality data, the best network training effect
was obtained from 35 (7×5) nerve cells, and the QE and TE values were 1.033 and 0.001, respectively.
When the average variance values are less than 5% for different clusters, the DBI is low; thus, the
corresponding clustering number can be regarded as the best clustering result. Therefore, this study input
the trained weights of the neuron nodes through the K-means cluster analysis combined with the DBI to
select the clustering number. The results are shown in Figure 3a. Six clusters were formed because this
number yielded the minimum DBI value (Figure 3a).
Fig. 3 (a) Davies–Bouldin index plot. (b) Results of the SOM clustering of the cells on the map plane (distribution of
the sampling sites on the SOM according to the eight water quality parameters and clustering of the trained SOM).
Figure 3b presents the results of the SOM clustering of the cells on the map plane, which exhibited
similarities among the different monitoring stations. In particular, Cluster 1 included the sampling points
around the Ganjia Lake Haloxylon natural conservation area in the southern Ebinur Lake region and points
east of Ebinur Lake and around the Kuitun River during the wet season. Cluster 2 included the monitoring
stations in the Jing River and around the agricultural ditch in the western Ebinur Lake region. Cluster 3
comprised samples of water from melted ice in the southern–western corner of the research zone, which
were grouped into only one type. Cluster 4 included the sampling points within the Ganjia Lake Haloxylon
natural conservation area during the dry season. Cluster 5 included the sampling points in the Jing River,
the agricultural ditch and around Ebinur Lake. Cluster 6 contained points located around the Kuitun River
and Ebinur Lake Bird Isle, which had more pools. In general, although individual points may interfere with
the explanation of the results, the classification results can identify the time sequence features in the
research zone. Clusters 1 to 3 entirely included samples from the wet season (May 2015), whereas Clusters
4 to 6 contained the monitoring samples from the dry season (October 2015). To further observe the
information on the water quality parameters of the Jinghe Oasis based on the response of different nerve
cells, the water quality information from the various cluster groups was visualized. The results are shown
in Figure 4.
Fig. 4 The patterning results for the water quality parameters on the SOM plane
5
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
Figure 4 shows the distribution of different water quality parameters on the SOM. High COD, TN,
NH3−N, and SD values were recorded in the right corner of the SOM network, thereby indicating a
declining trend in the southern Ebinur Lake region and the surrounding Kuitun River. The values during
the wet season were higher than those during the dry season. High values of SS and NUT were observed in
the left corner of the SOM network, which falls within the scope of the agricultural ditch and the Jing
River areas during the dry season. This region is mainly distributed around the agricultural land area and is
significantly influenced by human activities. High TP values were observed within the scope of the lower
left corner, which mainly focuses on the agricultural ditch and Jing River during the dry season. Crops
were mature during the dry season, and the farmland area was increased, which thereby increased the TP
value. In contrast, the distribution of the BOD5 was different, indicating a declining trend from the centre
to the surrounding area. High BOD5 values were mainly observed downstream of the Ganjia Lake
Haloxylon natural conservation zone, which is surrounded by a salt field, thereby exerting a certain
influence on the surrounding water quality. The regional change in the water quality of the Jinghe Oasis
was reflected clearly and directly through the SOM. To observe the distribution of the water quality
parameters directly, we collected different values of the water quality parameters at various layers (Figure
5).
Fig. 5 Average values for the water quality parameters
Figure 5 shows that the distribution of water quality parameters varies in different clustering layers.
Among the six clusters, water quality was generally relatively better in Cluster 3. However, Cluster 3 only
had one sampling point, which comprised ice and snow water. Therefore, its water quality was not
considered. In addition, the COD, SS, NUT, TN, and NH3−N contents were high in Clusters 1 and 2, which
indicates a relatively lower water quality compared with the water downstream of the Ganjia Lake
Haloxylon natural conservation zone, the surrounding Ebinur Lake region, and the agricultural land. The
SD values were high in Clusters 1, 4, and 6. Meanwhile, high BOD5 values were mainly concentrated in
Clusters 4 and 6, around the Ganjia Lake Haloxylon natural conservation zone and Ebinur Lake Bird Isle.
Many pools in these clusters are, to a certain extent, influenced by the water area. The concentration
difference in the TP of each layer was minimal and considerably influenced by human activities,
particularly changes in the agricultural land area. Based on these results and Chinese Environmental
Quality Standards for Surface Water (GB3838-2002), we evaluated the grades of the water quality
parameters, including the COD, BOD5, TN, NH3−N, and TP, at different layers (Table 1).
Table 1 The classes of water quality parameters in each cluster
As shown in Table 1, according to Chinese Environmental Quality Standards for Surface Water (GB
3838-2002), clusters 1 to 6 did not satisfy the potable water quality level. Clusters 1 and 2 had an identical
water quality classification level, and their COD and TN contents were higher than the standard values. In
particular, the COD content exceeded level V. Meanwhile, the BOD5 and COD contents were excessive in
Clusters 4 and 5, respectively. In Cluster 6, both the COD and BOD5 contents were excessive, but the COD
content was higher.
2.2 Analysis of land use/cover type and its relation to water quality at different layers
In this study, historical data, high-resolution Google Earth images, and field survey data were selected
to verify that more than 100 pixels of each land cover type were used for the training data, and the
confusion matrix was used to verify the classification results. For the classification results obtained in May
6
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
and October 2015 (Figure 6), precision increased to 89.9750% and 86.2848%, respectively, and the kappa
coefficients were 0.8681 and 0.8184, respectively, based on the confusion matrix (Table 2), which indicates
an accurate classification result that satisfies the research requirements. Accordingly, ArcGIS was used to
establish a 1 km buffer zone around the water sampling points. The composition of land use/cover at
different layers was analysed according to the hierarchical results of the water quality parameters, and the
results are presented in Figure 7.
Fig. 6 The change in land use/cover of the Ebinur Lake area during the rainy (May) and dry (October) seasons in 2015 (a:
May; b: October)
Fig. 7 The area of land use/cover for each cluster
Table 2 The confusion matrix calculation with a maximum likelihood supervised classification
Figure 7 shows the land use/cover mode at different layers. In general, the salinization phenomenon
was serious across the entire research zone. Among the six clusters, Cluster 1 mainly included the sampling
points around the Ganjia Lake Haloxylon natural conservation area, the eastern Ebinur Lake region, and
the Kuitun River. The major land type in this cluster was forest-grass land. The monitoring site in Cluster 2
was in the irrigation ditch of the Jinghe Oasis and in the Jing River. In May, crops do not grow abundantly
in the research zone, and the major land type in this cluster was forest-grass land. Cluster 4 mainly
included the Ganjia Lake Haloxylon natural conservation area and the sampling points in Tuotuo Village
during the dry season, which include farmland and forest-grassland. The sampling points in Cluster 5 were
in the agricultural ditch, the Jing River, and the surrounding Ebinur Lake. The land types mainly included
forest-grassland and farmland, which was larger than the other land type. Cluster 6 was mainly located in
the Kuitun River and the surrounding Ebinur Lake Bird Isle. A considerable number of plants, such as reed,
grow in some of its pools. The percentages of the forest-grass land and salinized land within the 1 km
buffer were large, based on the actual conditions in the research zone and the distribution of the sampling
points. Based on these results, the correlation between land use/cover type and water quality parameters
from Clusters 1 to 6 was analysed. The results are presented in Table 3.
Table 3 The correlation coefficients between the land use/cover and water quality parameters in each cluster
In Cluster 1, the forest–grass land exhibited a negative correlation with the SS and NUT under the
significance level of 0.01, with coefficients reaching up to −0.710 and −0.724, respectively. At a
significance level of 0.05, the water body type exhibited an obvious positive correlation with the COD and
a negative correlation with the TN, with coefficients of 0.986 and −0.721, respectively. At a confidence
level of 0.01, the salinized land demonstrated a positive correlation with the NUT, with a coefficient of
0.756. In Cluster 2, the farmland presented a negative correlation with the COD and a positive correlation
with the NUT at a confidence level of 0.05, and the coefficients were −0.581 and 0.639, respectively.
Under the same conditions, the forest–grass land exhibited a positive correlation with the COD, and the
coefficient was 0.613. At a confidence level of 0.01, the water body type exhibited an evident negative
correlation with the SS and NUT, and the correlation coefficients were −0.983 and −0.990, respectively. In
Cluster 4, several water quality parameters were mainly influenced by the farmland, forest-grass land, and
salinized land types. At a confidence level of 0.05, the farmland exhibited a negative correlation with the
COD, with a coefficient of −0.652. At a confidence level of 0.01, the farmland demonstrated an evident
positive correlation with TP, and the coefficient is 0.872. At a confidence level of 0.05, salinized land
shows a clear negative correlation with TP and NUT, and the coefficients were −0.791 and −0.819,
respectively. In this layer, the others land type exhibited a positive correlation with the TP, with a
7
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
coefficient of 0.868. The sampling points in this layer were mainly located around Tuotuo Village, where
the influences of human activities are considerable; therefore, in Cluster 4, the correlation percentage of
the others land type with TP was high. In Cluster 5, farmland demonstrated an evident negative correlation
with the BOD5 at a 0.01 confidence level, with a correlation coefficient reaching up to −0.881. At a
confidence level of 0.05, the salinized land showed a positive correlation with the BOD5, with a correlation
coefficient of 0.774. In Cluster 6, the forest–grass land showed a clear negative correlation at a confidence
level of 0.01, and the correlation coefficient reached −0.884. At a confidence level of 0.05, the water area
exhibited a positive correlation, with a correlation coefficient of 0.980.
From the results of the comprehensive analysis, the farmland, forest–grass land, and salinized land
types have a considerable influence on the water quality parameters in the Jinghe Oasis. In Clusters 1, 2,
and 6, the size of the water area, to a certain extent, also influenced changes in the water quality parameters.
Given the unbalanced distribution of the sampling points at different layers, the effect of the land use/cover
composition on the water quality in the research zone varied, so it can only indicate influences to a certain
extent. Therefore, considering the actual conditions in Ebinur Lake, different land use/cover types and
water quality influences must be understood as a whole, and the correlation between the land use/cover
types and water quality in the Jinghe Oasis at different periods is further discussed.
2.3 Analysis of land use/cover changes in the Jinghe Oasis and their correlation with water quality at different seasons
The constituents of the land use/cover types at different seasons exerted diverse influences on the
water quality. Therefore, an analysis of the influences of land use/cover type changes on water quality was
conducted. The results are presented in Figure 8.
Fig. 8 RDA analyses of comprehensive land use/cover and water quality (a: wet season;b: dry season)
As shown in Figure 8, the farmland exhibited a negative correlation with the COD at a confidence
level of 0.01 during the wet season, with a correlation coefficient of −0.543. In contrast, it showed a
positive correlation with the NUT, with a correlation coefficient of 0.555. At a confidence level of 0.05, the
farmland demonstrated a negative correlation with the NH3−N, and the correlation coefficient was −0.461.
At a confidence level of 0.05, the forest-grass land showed a positive correlation with the BOD5 and TP,
with correlation coefficients of 0.470 and 0.518, respectively. They exhibited a negative correlation with
the SS and NUT, with correlation coefficients of −0.529 and −0.498, respectively. At a confidence level of
0.05, the salinized land demonstrated a positive correlation with the BOD5 and TP, with correlation
coefficients of −0.503 and 0.518, respectively. In contrast, it presented a negative correlation with the SS
and NUT, with correlation coefficients of 0.449 and 0.449, respectively. During the dry season, the
influence of the farmland on various water quality parameters evidently increased because of crop growth.
At a confidence level of 0.01, the farmland presented a clear negative correlation with the COD and an
evident positive correlation with the TP, with correlation coefficients of −0.620 and 0.616, respectively. At
a confidence level of 0.05, the farmland showed a positive correlation with the TN and a negative
correlation with the BOD5, NH3−N and SD, with correlation coefficients of 0.543, −0.495, −0.522, and
−0.526 for TN, BOD5, NH3−N and SD, respectively. At a confidence level of 0.01, the salinized land
demonstrated a negative correlation with the NUT and TP, and the correlation coefficients were −0.543 and
−0.603, respectively. In contrast, it presented a positive correlation with the BOD5 at a confidence level of
0.05 and a correlation coefficient of 0.522. Similarly, during the wet and dry seasons, the correlation
between the water body and others land types with the water quality parameters was small. Based on these
8
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
results, the influences of the various land use/cover types in the research zone on the water quality
parameters exhibited the following order, in descending order of influence: farmland → forest–grass land
→ salinized land → water body → others. Moreover, the influence was lower during the wet season than
during the dry season.
3 Discussion
Given the seasonal differences, the unbalanced distribution of precipitation resulted in an apparent
variation in the surface runoff and further imbalanced the spatial distribution of the water quality in the
research zone (Fan et al.,2012; Prathumratana et al.,2008; Li et al.,2015). During the wet season (May) in
the Jinghe Oasis, melted water from mountain ice and snow promote the flow in the Jing River, thereby
resulting in a significant increase in surface runoff leading to an improved water quality in the rainy season
compared to the dry season. During the dry season, the aquatic plants in rivers and lakes grow as the
temperature rises, which can, to a certain degree, absorb and purify part of the water quality parameters.
Therefore, significant and seasonal changes in the surface runoff at the research zone are important factors
resulting in noticeable differences in the spatial distribution of water quality characteristics during the wet
and dry seasons. Another major factor that causes differences in the spatial distribution of water quality is
the change in land use/cover, especially farmland. During the dry season, farmland areas have a greater
influence on more water quality variables than they do during the rainy season because of their intensive
fertilization and agricultural runoff from soil erosion (Ngoye et al, 2004;Li et al., 2009; Tran et al., 2010).
Multiple factors threaten the ecological safety of the Jinghe Oasis system. Especially in recent years, the
lakeside desertification zone has rapidly expanded because of the decrease in the Ebinur Lake area and the
degradation of lakeside vegetation under influence of the strong winds in Alashankou. In the current
overall situation, the human activities influencing land use/cover changes are directly related to the
development of the vulnerable ecological area that surrounds Ebinur Lake.
Recent statistics indicate that the annual growth rate of the population in the Jinghe Oasis is
approximately 2.49%, which is slightly higher than previous growth rates (Li, 2006). Under the stress of a
large population, the amount of inappropriate activities that negatively impact land use/cover in the Jinghe
Oasis will increase. For the last 30 years, cotton has been the major crop in the Jinghe Oasis. The results of
the current study indicate that the sampling points surrounding the farmland in the research zone have
lower water quality values than the other studied regions. The primary livelihoods of the urban residents
around the Ebinur Lake area are agricultural and animal husbandry industries. Pollutants that result in high
TP and NH3−N contents in water include the excessive application of chemical fertilizers on farmlands, the
production of livestock manure in rural villages, randomly stocked garbage, and domestic wastewater. The
improper application of chemical fertilizers and pesticides to a vast area leads to high water nitrogen and
phosphorus contents, resulting in the spread of algae in river sections. Consequently, the amount of
dissolved oxygen in the river may decrease, the water quality may deteriorate, and eutrophication may
occur. Furthermore, this scenario poses a serious salinization problem. Certain measures have been
implemented for the ecological protection of Ebinur Lake, such as returning farmland to forest, cultivating
ecological forests, and promoting efficient irrigation and water-saving technologies. However, these
measures promote the gradual expansion of the lake area and result in different degrees of negative
consequences. The most apparent result has been the rise of the underground water level, which has
aggravated land salinization in the lowland areas and resulted in vast expanses of uncultivated former
agricultural lands. Statistics indicate that soil salinization in the Ebinur Lake area mainly occurs in Bortala
River, Jing River, the villages and towns surrounding Ebinur Lake, areas downstream of the Daheyanzi
9
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
River, and areas north of Bole City (Mi et al.,2010). Severe soil salinization has seriously affected the
farming of crops; therefore, some farmers have increased the amount of chemical fertilizers they apply to
increase yield, which also increases the pollution of the water and soil. Others have even abandoned their
land, thereby causing land use/cover change.
Most rivers in Xinjiang are characterized by a low water yield, short flow, small water environmental
capacity, poor self-cleaning capability, and low tolerance to pollution. Hence, an artificial change in the
land use and exploration of resources in lake regions lead to an evident correlation between land use/cover
types and water quality. In addition, the scientific utilization and protection of the water resources of
Ebinur Lake and the scientific application of chemical fertilizers and improvement of their application
rates are important actions and should be addressed to achieve sustainable development in the agricultural
irrigation zones of the Jinghe Oasis and rivers of Xinjiang.
4 Conclusions
The spatial distribution characteristics of water quality in the Jinghe Oasis and their correlation with
land use/cover types were analysed, and the following conclusions were drawn.
(1) Using the SOM method based on non-hierarchical k-means classification, 47 water quality
sampling points were divided into 6 types, and the time sequence characteristics of the research zone were
better recognized in the classification results. Clusters 1 to 3 comprised samples from the wet season (May
2015), whereas Clusters 4 to 6 comprised monitoring samples from the dry season (October 2015). In
general, the COD, SS, NUT, TN, and NH3−N contents were high. The SD value was high in Clusters 1, 4,
and 6. In addition, high BOD and TP values were mainly concentrated in Clusters 4 and 6. Based on these
findings, the water quality at different layers of the research zone was further evaluated. The results show
that Clusters 1 to 6 do not satisfy potable water quality standards.
(2) The correlations between the land use/cover types and water quality parameters for Clusters 1 to 6
were analysed according to the hierarchical results of the water quality parameters. The comprehensive
analysis indicates that the farmland, forest–grass land, and salinized land exerted significant influences on
the water quality parameters of the Jinghe Oasis. In Clusters 1, 2, and 6, the size of the water area, to a
certain extent, also influenced changes in the water quality parameters.
(3) During the wet and dry seasons, the influences that various land use/cover types in the research
zone had on the water quality parameters exhibited the following order, in descending order of influence:
farmland → forest–grass land → salinized land → water body → others. Moreover, the influences were
lower during the wet season than during the dry season.
In general, the land use/cover type, area percentage, and water quality in the Jinghe Oasis
demonstrated apparent correlations. The results of this study can tentatively explain the relationship
between water quality and land use/cover types in different clusters by the SOM method. This work
provides new insight for further studies on the correlation between land use/cover and water quality in the
Jinghe Oasis, as well as a scientific reference for formulating regulations and control policies for the
spatial development and water environmental protection of the Jinghe Oasis.
Acknowledgements: Thanks to the National Meteorological Information Center data provided
meteorological data. The research was carried out with the financial support provided by the Natural
Science Foundation of Xinjiang Uygur Autonomous Region, China (2016D01C029); National Natural
10
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
Science Foundation of China (41361045), the Scientific and technological talent training program of
Xinjiang Uygur Autonomous Region (grant No. QN2016JQ0041).The authors wish to thank the referees
for providing helpful suggestions to improve this manuscript.
References
An, Y., Zou, Z., & Li, R. 2016. Descriptive characteristics of surface water quality in hong kong by a self-organising map.
International Journal of Environmental Research & Public Health, 13(1):115.
Bu H M, Meng W, Zhang Y, Wan J.2014. Relationships between land use patterns and water quality in the Taizi River basin,
China. Ecological Indicators, 41: 187-197.
Bierman P, Lewis M, Ostendorf B, Tanner J.2011.A review of methods for analysing spatial and temporal patterns in coastal
water quality. Ecological Indicators, 11(1): 103-114.
Céréghino R, Park Y S.2009. Review of the self-organizing map (SOM) approach in water resources: commentary.
Environmental Modelling & Software, 24(8): 945-947.
Chon T S. 2011. Self-organizing maps applied to ecological sciences. Ecological Informatics, 6(1): 50-61.
De Jonge M, Van de Vijver B, Blust R, Bervoets L.2008.Responses of aquatic organisms to metal pollution in a lowland
river in Flanders: a comparison of diatoms and macroinvertebrates. Science of the Total Environment, 407(1): 615-629.
Fan X Y, Cui B S, Zhang K J, Zhang Z M.2012.Water quality management based on division of dry and wet seasons in Pearl
River Delta, China. Clean-Soil,Air,Water, 40(4):381-393.
Huang J L, Li Q S, Huang L,Zhang Z F,Mu J L, Huang Y L.2013a. Watershed-scale evaluation for land-based nonpoint
source nutrients management in the Bohai Sea Bay, China. Ocean & coastal management, 71: 314-325.
Huang J L, Li Q S, Pontius Jr R G, Hong H S. 2013b. Detecting the dynamic linkage between landscape characteristics and
water quality in a subtropical coastal watershed, Southeast China. Environmental management, 51(1): 32-44.
Hur J, Nguyen H V M, Lee B M.2014. Influence of upstream land use on dissolved organic matter and trihalomethane
formation potential in watersheds for two different seasons. Environmental Science and Pollution Research, 21(12):
7489-7500.
Jilil A, Mu G J.2002.Analysis on the dust storms and their disasters in the lakebed region of Ebinur Lake,Xinjiang.Arid Land
Geography, 25(2):122-126.
Kohonen T. 2001.Self-Organizing Maps. Berlin: Springer.
Kohonen T.2013. Essentials of the self-organizing map. Neural Networks, 37: 52-65.
Kalteh A M, Hjorth P, Berndtsson R.2008.Review of the self-organizing map (SOM) approach in water resources: Analysis,
modelling and application. Environmental Modelling & Software, 23(7): 835-845.
Lee S W, Hwang S J, Lee S B,Sung H C. 2009. Landscape ecological approach to the relationships of land use patterns in
watersheds to water quality characteristics. Landscape and Urban Planning, 92(2): 80-89.
Li Y Y, Jiao J X, Wang Y,Yang W, Meng C, Li B Z, li Y, Wu J S. 2015. Characteristics of nitrogen loading and its influencing
factors in several typical agricultural watersheds of subtropical China. Environmental Science and Pollution Research,
22(3): 1831-1840.
Li W, Yao X Y, Liang Z W, Wu Y M, Shi J Y, Chen Y X. 2013. Assessment of surface water quality using self-organizing
map and Hasse diagram technique. Acta Scientiae Circumstantiae, 33 (3) :893-903.
Li J, Hao Z G.2010.Tourism planning based on Google Earth virtual earth platform.Remote Sensing for Land & Resources,
(1):130-133.
Li K,Wang L,Li Z H,Wang X R,Chen H B,Wu Z,Zhu P. 2015.Spatial variability characteristics of water quality and its
driving forces in honghu lake during high water-level period. Environmental Science, 36(4):1285-1292.
Li Y H. 2006.Water resources carrying capacity in the Ebinur Lake basin of Xinjiang.Shanghai:East China Normal
University.
Li S, Gu S, Tan X, Zhang Q.2009. Water quality in the upper Han River basin, China: the impacts of land use/land cover in
11
12
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
riparian buffer zone. Journal of Hazardous Materials, 165, 317–324.
Mi Y,Chang S L,Shi Q D,Gao X,Huang C.2010.Study on the effect of agricultural non-point source pollution to water
environment of the Ebinur Lake basin during high flow period. Arid Zone Research, 27(2):278-283.
Ngoye E, Machiwa J F.2004.The influence of land use patterns in the Ruvu river watershed on water quality in the river
system. Phys. Chem. Earth, 29,1161-1166.
National Research Council.2008. Integrating Multiscale Observations of U.S. Waters; The National Academies Press:
Washington, DC, USA.
Park Y S, Kwon Y S, Hwang S J, Park S. 2014.Characterizing effects of landscape and morphometric factors on water
quality of reservoirs using a self-organizing map. Environmental Modelling & Software, 55: 214-221.
Prathumratana L, Sthiannopkao S, Kim K W.2008. The relationship of climatic and hydrological parameters to surface water
quality in the lower Mekong River. Environment International, 34(6): 860-866.
Swaney D P, Humborg C, Emeis K, Kannen A, Silvert W, Pastres R, Solidoro C, Yamamuro M, Hénocque Y, Nicholls R.
2012. Five critical questions of scale for the coastal zone. Estuarine, Coastal and Shelf Science, 96: 9-21.
Shen Z, Hou X, Li W, Aini G, Chen L, Gong Y. 2015. Impact of landscape pattern at multiple spatial scales on water quality:
A case study in a typical urbanised watershed in China. Ecological Indicators, 48: 417-427.
Sun F, Sun W, Chen J, Gong P. 2012. Comparison and improvement of methods for identifying waterbodies in remotely
sensed imagery. International journal of remote sensing 33(21),6854-6875.
Tran C P, Bode R W, Smith A J, Kleppel G S.2010.Land-use proximity as a basis for assessing stream water quality in New
York State (USA). Ecological Indicators, 10, 727–733.
Ter Braak C J F, Smilauer P. 2002. CANOCO reference manual and CanoDraw for Windows user's guide: software for
canonical community ordination (version 4.5). In: Section on Permutation Methods. Microcomputer Power, Ithaca, NY.
Uuemaa E, Roosaare J, Mander Ü.2005.Scale dependence of landscape metrics and their indicatory value for nutrient and
organic matter losses from catchments. Ecological Indicators, 5(4): 350-369.
Wan R, Cai S, Li H,Yang G, Li Z, Nie X. 2014. Inferring land use and land cover impact on stream water quality using a
Bayesian hierarchical modeling approach in the Xitiaoxi River Watershed, China. Journal of Environmental
Management, 133: 1-11.
Wang XP, Zhang F, Li XH, Cao C, Guo M. 2017. Correlation analysis between the spatial characteristics of land
use/cover-landscape pattern and surface-water quality in the Ebinur Lake area. Acta Ecologica sinica
2017,37(22):7438~7452.(In Chinese)
Xiao H, Ji W.2007. Relating landscape characteristics to non-point source pollution in mine waste-located watersheds using
geospatial techniques. Journal of Environmental Management, 82(1): 111-119.
Yang Y N,Wang J L, Cheng G J, Xi X H, Wang C.2016.Relationship between land use pattern and water quality change in
Fuxian Lake basin. Remote Sensing for Land and Resources, 28(1):159-165.
Zhang, Y., Liang, X., Wang, Z., & Xu, L. 2015. A novel approach combining self-organizing map and parallel factor analysis
for monitoring water quality of watersheds under non-point source pollution. Scientific Reports, 5(3), 16079.
Zhou P, Huang J L, Pontius Jr R G, Hong H S. 2016. New insight into the correlations between land use and water quality in
a coastal watershed of China: Does point source pollution weaken it? Science of the Total Environment, 543: 591-600.
Zhang F, Tiyip T, Johnson V C, Ding J L, Zhou M,Chan N W. 2015. The influence of natural and human factors in the
shrinking of the Ebinur Lake, Xinjiang, China, during the 1972–2013 period. Environmental Monitoring and
Assessment, 187(1): 1-14.
Figure captions
Fig. 1 Location of the study area
Fig. 2 Conceptual model of the methodology
13
(a)
0.4
0.6
0.8
1.0
1.2
1.4
1.6
2 3 4 5 6 7
Clusters
Dav
ies-
Bou
ldin
inde
x
Fig. 3 (a) Davies–Bouldin index plot. (b) Results of the SOM clustering of the cells on the map plane (distribution of the
sampling sites on the SOM according to the eight water quality parameters and clustering of the trained SOM).
Fig. 4 The patterning results for the water quality parameters on the SOM plane
14
Fig. 5 Average values for the water quality parameters
(b) (a)
Fig. 6 The change in land use/cover of the Ebinur Lake area during the rainy (May) and dry (October) seasons in 2015 (a: May; b:
October)
Fig. 7 The area of land use/cover for each cluster
15
(b) (a)
Fig. 8 RDA analyses of comprehensive land use/cover and water quality (a: wet season;b: dry season)
16
17
Tables caption
Table 1 The classes of water quality parameter in each cluster
Cluster1 Cluster2 Cluster4 Cluster5 Cluster6
COD ExceedⅤ ExceedⅤ Ⅰ Ⅳ Ⅴ
BOD5 Ⅰ Ⅰ Ⅳ Ⅰ Ⅳ
TN Ⅳ Ⅳ Ⅲ Ⅲ Ⅲ
NH3-N Ⅱ Ⅱ Ⅰ Ⅰ Ⅰ
TP Ⅱ Ⅱ Ⅱ Ⅲ Ⅱ
Table 2 The calculation of a confusion matrix by a maximum likelihood supervised classification
LULC Water body Saline land Farmland Forest
grassland
Other
land Total
User’s
accuracy
(%)
Water body 144 0 0 0 0 144 100
Saline land 0 77 0 0 16 93 82.79
Farmland 0 36 101 0 0 137 73.72
Forest-Grass land 0 36 0 101 0 137 73.72
Other land types 1 0 0 0 87 88 98.96
Total 145 149 101 101 103 Overall=89.9750%
May
Producer’s accuracy (%) 99.31 51.67 100 100 84.46 Kappa=0.8681
Water body 144 0 0 0 0 144 100
Saline land 0 57 0 0 26 83 86.67
Farmland 0 16 101 0 0 117 86.32
Forest-Grass land 4 16 0 101 0 117 86.32
Other land types 0 0 0 0 77 77 100
Total 148 89 101 101 103 Overall=86.2848%
October
Producer’s accuracy (%) 97.29 64 100 100 74.75 Kappa=0.8184
Table 3 The correlation coefficients between land use/cover and water quality parameters in each cluster
Parameters Farmland Forest-Grass land Water Body Salinized land Others
COD -0.161 0.240 0.986* -0.110 -0.361
BOD5 0.074 0.492 -0.439 -0.613 0.552
SS -0.271 -0.710** 0.801 0.619 -0.384
TP -0.195 0.453 0.371 0.444 0.623
TN 0.464 0.524 -0.721* -0.224 0.121
NH3-N -0.491 0.039 0.071 -0.066 0.291
SD -0.296 0.448 0.415 -0.426 0.396
Cluster1
NUT -0.261 -0.724** 0.550 0.756** -0.612
COD -0.581* 0.613* 0.916 -0.693** 0.442
BOD5 -0.004 0.455 0.055 -0.545 0.242
SS 0.493 -0.512 -0.983** 0.386 0.047
Cluster2
TP -0.222 0.531 0.850 -0.129 0.382
18
TN 0.351 0.415 -0.867 -0.356 -0.311
NH3-N -0.467 0.121 0.122 -0.269 0.284
SD -0.226 -0.073 -0.051 -0.217 0.473
NUT 0.639* -0.446 -0.990** 0.513 -0.236
COD -0.652* 0.484 / 0.375 -0.048
BOD5 -0.482 -0.402 / 0.505 0.688
SS -0.155 0.658 / -0.179 -0.167
TP 0.872** -0.398 / -0.791* 0.868*
TN 0.336 0.468 / -0.571 -0.124
NH3-N -0.202 -0.540 / 0.398 0.352
SD -0.543 0.825* / 0.214 -0.549
Cluster4
NUT 0.578 0.469 / -0.819* -0.129
COD 0.094 -0.372 / 0.325 -0.400
BOD5 -0.881** 0.503 / 0.774* -0.044
SS 0.621 -0.533 / -0.284 -0.380
TP 0.587 -0.702 / -0.565 0.136
TN 0.735 -0.588 / -0.184 -0.604
NH3-N -0.675 0.108 / 0.487 0.308
SD -0.632 0.330 / 0.208 0.576
Cluster5
NUT 0.311 0.076 / -0.459 0.154
COD 0.489 0.401 0.980* -0.454 0.289
BOD5 -0.256 -0.884** -0.660 0.367 0.327
SS -0.481 0.194 0.341 -0.150 -0.062
TP -0.545 -0.656 0.269 -0.060 0.206
TN -0.158 -0.364 -0.022 0.516 0.313
NH3-N -0.553 -0.366 0.517 -0.090 0.603
SD 0.811 -0.037 -0.857 0.249 -0.282
Cluster6
NUT 0.450 0.165 0.764 -0.497 0.636
*p<0.05(2-tailed) ** p<0.01(2-tailed)